Check your topic replication factors
Kafka services rely on replication between brokers to preserve data in case of the loss of a node. Consider how business critical the data in each topic is and make sure that replication is set high enough for it.
You can set this in the Aiven web console when you create a new topic or edit an existing one.
Note: To prevent data loss from unexpected node termination, you cannot set the replication factor below 2 in the Aiven web console.
Choose a reasonable number of partitions for a topic
Too few partitions can cause bottlenecks in data processing. In the most extreme case, a single partition means that messages are effectively processed sequentially. However, too many partitions causes strain on the cluster because of the additional overhead. As you cannot reduce the number of partitions for existing topics, it is usually best to start with a low number that allows efficient data processing and increase it if needed.
As a general rule of thumb, the recommendation is to have max 4000 partitions per broker, and max 200 000 partitions per cluster.
Periodically examine topics that use entity-based partitioning for imbalances
If you partition messages based on an entity ID (for example, user ID), there is a risk of heavily imbalanced partitions. This results in uneven load in your cluster and reduces how effectively it can process messages in parallel.
You can check this in the Aiven web console when you click on a topic in the list view on the Topics tab of your service.
Find the right balance between throughput and latency
Try different batch sizes in your producer and consumer configurations to find the right balance. Bigger batches increase throughput but also increase the latency for individual messages. Conversely, using smaller batches decreases message processing latency, but the overhead per message increases and the overall throughput decreases.
Pay attention to
tasks.max for connector configs
By default, connectors run a maximum of
1 task, which usually leads to underutilization for large Kafka Connect services (unless you have many connectors with one task each). In general, it is good to keep the cluster CPUs occupied with connector tasks without overloading them. If your connector is under-performing, you can try increasing
tasks.max to match the number of partitions.
Consider a standalone Kafka Connect service
You can run Kafka Connect as part of your existing Aiven for Apache Kafka service (for business-4 and higher service plans). While this allows you to try out Kafka Connect by enabling the feature within your existing service, for heavy usage we recommend that you enable a standalone Kafka Connect service to run your connectors. This allows you to scale the Kafka service and the connector service independently and offers more CPU time and memory for the Kafka Connect service.